70322ae3 |
1 | /* |
2 | * du.c: implementation of du.h. |
3 | */ |
4 | |
dcf1fcc0 |
5 | #include "agedu.h" /* for config.h */ |
6 | |
7 | #ifdef HAVE_FEATURES_H |
70322ae3 |
8 | #define _GNU_SOURCE |
9 | #include <features.h> |
dcf1fcc0 |
10 | #endif |
70322ae3 |
11 | |
12 | #include <stdio.h> |
13 | #include <string.h> |
14 | #include <stdlib.h> |
15 | #include <errno.h> |
16 | |
17 | #include <sys/types.h> |
18 | #include <sys/stat.h> |
19 | #include <unistd.h> |
20 | |
21 | #include "du.h" |
995db599 |
22 | #include "alloc.h" |
70322ae3 |
23 | |
9c6e61f2 |
24 | #if !defined __linux__ || defined HAVE_FDOPENDIR |
704fafa3 |
25 | |
26 | /* |
9c6e61f2 |
27 | * Wrappers around POSIX opendir, readdir and closedir, which |
28 | * permit me to replace them with different wrappers in special |
29 | * circumstances. |
704fafa3 |
30 | */ |
31 | |
9c6e61f2 |
32 | #include <dirent.h> |
33 | typedef DIR *dirhandle; |
704fafa3 |
34 | |
35 | int open_dir(const char *path, dirhandle *dh) |
36 | { |
9c6e61f2 |
37 | #if defined O_NOATIME && defined HAVE_FDOPENDIR |
38 | |
39 | /* |
40 | * On Linux, we have the O_NOATIME flag. This means we can |
41 | * read the contents of directories without affecting their |
42 | * atimes, which enables us to at least try to include them in |
43 | * the age display rather than exempting them. |
44 | * |
45 | * Unfortunately, opendir() doesn't let us open a directory |
46 | * with O_NOATIME. So instead, we have to open the directory |
47 | * with vanilla open(), and then use fdopendir() to translate |
48 | * the fd into a POSIX dir handle. |
49 | */ |
50 | int fd; |
51 | |
52 | fd = open(path, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_LARGEFILE | |
53 | O_NOATIME | O_DIRECTORY); |
54 | if (fd < 0) { |
704fafa3 |
55 | /* |
56 | * Opening a file with O_NOATIME is not unconditionally |
57 | * permitted by the Linux kernel. As far as I can tell, |
58 | * it's permitted only for files on which the user would |
59 | * have been able to call utime(2): in other words, files |
60 | * for which the user could have deliberately set the |
61 | * atime back to its original value after finishing with |
62 | * it. Hence, O_NOATIME has no security implications; it's |
63 | * simply a cleaner, faster and more race-condition-free |
64 | * alternative to stat(), a normal open(), and a utimes() |
65 | * when finished. |
66 | * |
67 | * The upshot of all of which, for these purposes, is that |
68 | * we must be prepared to try again without O_NOATIME if |
69 | * we receive EPERM. |
70 | */ |
71 | if (errno == EPERM) |
9c6e61f2 |
72 | fd = open(path, O_RDONLY | O_NONBLOCK | O_NOCTTY | |
73 | O_LARGEFILE | O_DIRECTORY); |
74 | if (fd < 0) |
75 | return -1; |
76 | } |
77 | |
78 | *dh = fdopendir(fd); |
79 | #else |
80 | *dh = opendir(path); |
81 | #endif |
82 | |
83 | if (!*dh) |
84 | return -1; |
85 | return 0; |
86 | } |
87 | |
88 | const char *read_dir(dirhandle *dh) |
89 | { |
90 | struct dirent *de = readdir(*dh); |
91 | return de ? de->d_name : NULL; |
92 | } |
93 | |
94 | void close_dir(dirhandle *dh) |
95 | { |
96 | closedir(*dh); |
97 | } |
98 | |
99 | #else /* defined __linux__ && !defined HAVE_FDOPENDIR */ |
100 | |
101 | /* |
102 | * Earlier versions of glibc do not have fdopendir(). Therefore, |
103 | * if we are on Linux and still wish to make use of O_NOATIME, we |
104 | * have no option but to talk directly to the kernel system call |
105 | * interface which underlies the POSIX opendir/readdir machinery. |
106 | */ |
107 | |
108 | #define __KERNEL__ |
109 | #include <unistd.h> |
110 | #include <fcntl.h> |
111 | #include <linux/types.h> |
112 | #include <linux/dirent.h> |
113 | #include <linux/unistd.h> |
114 | |
115 | _syscall3(int, getdents, uint, fd, struct dirent *, dirp, uint, count) |
116 | |
117 | typedef struct { |
118 | int fd; |
119 | struct dirent data[32]; |
120 | struct dirent *curr; |
121 | int pos, endpos; |
122 | } dirhandle; |
123 | |
124 | int open_dir(const char *path, dirhandle *dh) |
125 | { |
126 | /* |
127 | * As above, we try with O_NOATIME and then fall back to |
128 | * trying without it. |
129 | */ |
130 | dh->fd = open(path, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_LARGEFILE | |
131 | O_NOATIME | O_DIRECTORY); |
132 | if (dh->fd < 0) { |
133 | if (errno == EPERM) |
134 | dh->fd = open(path, O_RDONLY | O_NONBLOCK | O_NOCTTY | |
135 | O_LARGEFILE | O_DIRECTORY); |
704fafa3 |
136 | if (dh->fd < 0) |
137 | return -1; |
138 | } |
139 | |
140 | dh->pos = dh->endpos = 0; |
141 | |
142 | return 0; |
143 | } |
144 | |
145 | const char *read_dir(dirhandle *dh) |
146 | { |
147 | const char *ret; |
148 | |
149 | if (dh->pos >= dh->endpos) { |
150 | dh->curr = dh->data; |
151 | dh->pos = 0; |
152 | dh->endpos = getdents(dh->fd, dh->data, sizeof(dh->data)); |
153 | if (dh->endpos <= 0) |
154 | return NULL; |
155 | } |
156 | |
157 | ret = dh->curr->d_name; |
158 | |
159 | dh->pos += dh->curr->d_reclen; |
160 | dh->curr = (struct dirent *)((char *)dh->data + dh->pos); |
161 | |
162 | return ret; |
163 | } |
164 | |
165 | void close_dir(dirhandle *dh) |
166 | { |
167 | close(dh->fd); |
168 | } |
169 | |
9c6e61f2 |
170 | #endif /* !defined __linux__ || defined HAVE_FDOPENDIR */ |
70322ae3 |
171 | |
172 | static int str_cmp(const void *av, const void *bv) |
173 | { |
174 | return strcmp(*(const char **)av, *(const char **)bv); |
175 | } |
176 | |
177 | static void du_recurse(char **path, size_t pathlen, size_t *pathsize, |
178 | gotdata_fn_t gotdata, void *gotdata_ctx) |
179 | { |
704fafa3 |
180 | const char *name; |
70322ae3 |
181 | dirhandle d; |
9c6e61f2 |
182 | STRUCT_STAT st; |
70322ae3 |
183 | char **names; |
184 | size_t i, nnames, namesize; |
185 | |
9c6e61f2 |
186 | if (LSTAT(*path, &st) < 0) { |
70322ae3 |
187 | fprintf(stderr, "%s: lstat: %s\n", *path, strerror(errno)); |
188 | return; |
189 | } |
190 | |
191 | if (!gotdata(gotdata_ctx, *path, &st)) |
192 | return; |
193 | |
194 | if (!S_ISDIR(st.st_mode)) |
195 | return; |
196 | |
197 | names = NULL; |
198 | nnames = namesize = 0; |
199 | |
704fafa3 |
200 | if (open_dir(*path, &d) < 0) { |
70322ae3 |
201 | fprintf(stderr, "%s: opendir: %s\n", *path, strerror(errno)); |
202 | return; |
203 | } |
704fafa3 |
204 | while ((name = read_dir(&d)) != NULL) { |
205 | if (name[0] == '.' && (!name[1] || (name[1] == '.' && !name[2]))) { |
206 | /* do nothing - we skip "." and ".." */ |
207 | } else { |
208 | if (nnames >= namesize) { |
209 | namesize = nnames * 3 / 2 + 64; |
210 | names = sresize(names, namesize, char *); |
70322ae3 |
211 | } |
704fafa3 |
212 | names[nnames++] = dupstr(name); |
213 | } |
70322ae3 |
214 | } |
704fafa3 |
215 | close_dir(&d); |
70322ae3 |
216 | |
217 | if (nnames == 0) |
218 | return; |
219 | |
220 | qsort(names, nnames, sizeof(*names), str_cmp); |
221 | |
222 | for (i = 0; i < nnames; i++) { |
223 | size_t newpathlen = pathlen + 1 + strlen(names[i]); |
224 | if (*pathsize <= newpathlen) { |
225 | *pathsize = newpathlen * 3 / 2 + 256; |
226 | *path = sresize(*path, *pathsize, char); |
227 | } |
256c29a2 |
228 | /* |
229 | * Avoid duplicating a slash if we got a trailing one to |
230 | * begin with (i.e. if we're starting the scan in '/' itself). |
231 | */ |
232 | if (pathlen > 0 && (*path)[pathlen-1] == '/') { |
233 | strcpy(*path + pathlen, names[i]); |
234 | newpathlen--; |
235 | } else { |
236 | sprintf(*path + pathlen, "/%s", names[i]); |
237 | } |
70322ae3 |
238 | |
239 | du_recurse(path, newpathlen, pathsize, gotdata, gotdata_ctx); |
240 | |
241 | sfree(names[i]); |
242 | } |
243 | sfree(names); |
244 | } |
245 | |
444c684c |
246 | void du(const char *inpath, gotdata_fn_t gotdata, void *gotdata_ctx) |
70322ae3 |
247 | { |
248 | char *path; |
249 | size_t pathlen, pathsize; |
250 | |
251 | pathlen = strlen(inpath); |
252 | pathsize = pathlen + 256; |
253 | path = snewn(pathsize, char); |
254 | strcpy(path, inpath); |
255 | |
256 | du_recurse(&path, pathlen, &pathsize, gotdata, gotdata_ctx); |
257 | } |